Guangzhou Laboratory Academic Exchange: Dr. Tianyuan Wang Shares AI Antibody Design R&D Practice

Obtaining candidate antibodies that recognize a target epitope has always been a critical step in antibody drug R&D. Dr. Wang introduced the direction the team has been continuously exploring: directly generating candidate antibodies around a target epitope using AI. This approach has now been applied to multiple projects including cytokine receptors and ADC targets, completing the full workflow from computational design to molecular binding validation and cell functional validation — forming a closed-loop R&D cycle that combines computational design with experimental verification.

Two antibody engineering optimization project cases were shared. One case focused on a GPCR agonistic nanobody, where AI optimization enhanced antibody activity. The other case targeted a CAR-T antibody, improving its specific binding to mutants and providing new ideas for improving the therapeutic window.

Reflecting on the current state of AI antibody design, Dr. Wang stated that the core challenge still constraining further development of AI de novo antibody design is the shortage of high-quality data, which directly impacts model generalizability. Compared to ordinary proteins, antibody structure prediction is more complex. High-quality antigen–antibody complex structural data is difficult and costly to obtain, limiting the effective data available for model training and affecting design success rates. To address this challenge, Clickmab is leveraging high-throughput, low-cost data acquisition technology to continuously build paired datasets of antibody sequences and antigen epitopes (Patches). The plan is to accumulate millions of high-quality Patch–Sequence pairs over the next two years, continuously optimizing model capability and substantially improving the success rate of AI de novo antibody design.